Update inference perf chart to match inf perf repo.

rlakhtakia · rlakhtakia · commit 142adc09132d · 2025-12-04T11:13:30.000Z
diff --git a/benchmarking/benchmark-values.yaml b/benchmarking/benchmark-values.yaml
@@ -1,9 +1,9 @@
 job:
   image:
     repository: quay.io/inference-perf/inference-perf
-    tag: "latest" # Defaults to .Chart.AppVersion
-  serviceAccountName: ""
+    tag: "" # Defaults to .Chart.AppVersion
   nodeSelector: {}
+  serviceAccountName: ""
   # Example resources:
   # resources:
   #   requests:
@@ -18,19 +18,27 @@ logLevel: INFO
 
 # A GCS bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/gcs-dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/gcs-dataset.json.
+# at /gcsDataset/gcs-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /gcsDataset/gcs-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 gcsPath: ""
 
-# A S3 bucket path that points to the dataset file.
+# An S3 bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/s3-dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/s3-dataset.json.
+# at /s3Dataset/s3-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /s3Dataset/s3-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 s3Path: ""
 
-# hfToken optionally creates a secret with the specified token.
-# Can be set using helm install --set hftoken=<token>
-hfToken: ""
+# Optional Token configuration for Hugging Face authentication.
+# hfSecret: Configures a pre-existing Kubernetes Secret.
+# hfToken: Creates a new kubernetes secret with the specified token.
+# If both specified, 'hfSecret' takes precedence over 'hfToken'.
+token:
+  hfSecret:
+    name: "" # The name of the existing Secret (e.g., 'my-hf-secret').
+    key: ""  # The key within the Secret that holds the token value (e.g., 'token' or 'hf-token').
+  hfToken: ""
 
 config:
   load:
diff --git a/benchmarking/inference-perf/Chart.yaml b/benchmarking/inference-perf/Chart.yaml
@@ -3,4 +3,4 @@ name: inference-perf
 description: A Helm chart for running inference-perf benchmarking tool
 type: application
 version: 0.2.0
-appVersion: "0.2.0"
+appVersion: "v0.2.0"
diff --git a/benchmarking/inference-perf/README.md b/benchmarking/inference-perf/README.md
@@ -19,11 +19,22 @@ Make sure you have the following tools installed and configured:
 
 Before deployment, navigate to the **`deploy/inference-perf`** directory and edit the **`values.yaml`** file to customize your deployment and the benchmark parameters.
 
-#### Optional Parameters
+#### Optional Token Parameters
+Hugging Face token can be provided either by providing a value (`hfToken`) or by referencing an existing Kubernetes Secret (`hfSecret.Name` and `hfSecret.Key`).
+
+> If both `hfToken` and the `hfSecret` parameters are provided, the chart logic is configured to prioritize the `hfSecret` reference.
 
 | Key | Description | Default |
 | :--- | :--- | :--- |
 | `hfToken` | Hugging Face API token. If provided, a Kubernetes `Secret` named `hf-token-secret` will be created for authentication. | `""` |
+| `hfSecret.name` | The name of a pre-existing Kubernetes Secret that contains a Hugging Face API token. | `""` |
+| `hfSecret.key` | The key within the pre-existing Kubernetes Secret that holds the token value. | `""` |
+---
+
+#### Optional Job Parameters
+
+| Key | Description | Default |
+| :--- | :--- | :--- |
 | `serviceAccountName` | Standard Kubernetes `serviceAccountName`. If not provided, default service account is used. | `""` |
 | `nodeSelector` |  Standard Kubernetes `nodeSelector` map to constrain pod placement to nodes with matching labels. | `{}` |
 | `resources` | Standard Kubernetes resource requests and limits for the main `inference-perf` container. | `{}` |
@@ -54,7 +65,29 @@ The identity executing the workload (e.g., the associated Kubernetes Service Acc
 
 | Key | Description | Default |
 | :--- | :--- | :--- |
-| `gcsPath` | A GCS URI pointing to the dataset file (e.g., `gs://my-bucket/dataset.json`). The file will be automatically copied to the running pod during initialization. | `""` |
+| `gcsPath` | A GCS bucket name pointing to the dataset file (e.g., `<my-bucket-path-to-file>/dataset.json`). The file will be automatically copied to the running pod during initialization. The file will be copied to `gcsDataset/dataset.json` | `""` |
+
+---
+
+#### AWS Specific Parameters
+
+This section details the necessary configuration and permissions for using an S3 path to manage your dataset, typical for deployments on AWS EKS.
+
+##### Required IAM Permissions
+
+The identity executing the workload (e.g., the associated Kubernetes Service Account, often configured via IRSA - IAM Roles for Service Accounts) must possess an associated AWS IAM Policy that grants the following S3 Actions on the target S3 bucket for data transfer:
+
+* **S3 Read/Download (Object Access)**
+    * Action: `s3:GetObject` (Required to download the input dataset from S3).
+    * Action: `s3:ListBucket` (Often required to check for the file's existence and list bucket contents).
+
+* **S3 Write/Upload (Object Creation)**
+    * Action: `s3:PutObject` (Required to upload benchmark results back to S3).
+
+
+| Key | Description | Default |
+| :--- | :--- | :--- |
+| `s3Path` | An S3 bucket name pointing to the dataset file (e.g., `<my-bucket-path-to-file>/dataset.json`). The file will be automatically copied to the running pod during initialization. The file will be copied to `s3Dataset/dataset.json` | `""` |
 
 ---
 
@@ -80,6 +113,6 @@ Use the **`helm install`** command from the **`deploy/inference-perf`** director
 ### 4. Cleanup
 
 To remove the benchmark deployment.
-    ```bash
+```bash
     helm uninstall test
-    ```
+```
diff --git a/benchmarking/inference-perf/templates/job.yaml b/benchmarking/inference-perf/templates/job.yaml
@@ -23,19 +23,19 @@ spec:
       initContainers:
         - name: fetch-gcs-dataset
           image: google/cloud-sdk:latest
-          command: ["sh", "-c", "gsutil cp {{ .Values.gcsPath }} /dataset/gcs-dataset.json"]
+          command: ["sh", "-c", "gsutil cp gs://{{ .Values.gcsPath }} /gcsDataset/gcs-dataset.json"]
           volumeMounts:
-            - name: dataset-volume
-              mountPath: /dataset
+            - name: gcs-dataset-volume
+              mountPath: /gcsDataset
 {{- end }}
 {{- if .Values.s3Path}}
       initContainers:
         - name: fetch-s3-dataset
           image: google/cloud-sdk:latest
-          command: ["sh", "-c", "aws s3 cp s3://{{ .Values.s3Path }} /dataset/s3-dataset.json"]
+          command: ["sh", "-c", "aws s3 cp s3://{{ .Values.s3Path }} /s3Dataset/s3-dataset.json"]
           volumeMounts:
-            - name: dataset-volume
-              mountPath: /dataset
+            - name: s3-dataset-volume
+              mountPath: /s3dataset
 {{- end }}
       containers:
         - name: inference-perf-container
@@ -47,20 +47,42 @@ spec:
             - "--log-level"
             - {{ .Values.logLevel }}
           env:
-            {{- if .Values.hfToken }}
+{{- if and .Values.token.hfSecret.name .Values.token.hfSecret.key }}
+            - name: HF_TOKEN
+              valueFrom:
+                secretKeyRef:
+                  name: {{ .Values.token.hfSecret.name }}
+                  key: {{ .Values.token.hfSecret.key }}
+{{- else if .Values.token.hfToken }}
             - name: HF_TOKEN
               valueFrom:
                 secretKeyRef:
                   name: {{ include "inference-perf.hfSecret" . }}
                   key: {{ include "inference-perf.hfKey" . }}
-            {{- end }}
+{{- end }}
           volumeMounts:
             - name: config-volume
               mountPath: {{ include "inference-perf.configMount" . }}
               readOnly: true
+{{- if .Values.gcsPath}}
+            - name: gcs-dataset-volume
+              mountPath: /gcsDataset
+{{- end }}
+{{- if .Values.s3Path}}
+            - name: s3-dataset-volume
+              mountPath: /s3Dataset
+{{- end }}
           resources:
             {{- toYaml .Values.job.resources | nindent 12 }}
       volumes:
         - name: config-volume
           configMap:
             name: {{ include "inference-perf.fullname" . }}-config
+{{- if .Values.gcsPath}}
+        - name: gcs-dataset-volume
+          emptyDir: {}
+{{- end }}
+{{- if .Values.s3Path}}
+        - name: s3-dataset-volume
+          emptyDir: {}
+{{- end }}
diff --git a/benchmarking/prefix-cache-aware/high-cache-values.yaml b/benchmarking/prefix-cache-aware/high-cache-values.yaml
@@ -2,9 +2,9 @@
 job:
   image:
     repository: quay.io/inference-perf/inference-perf
-    tag: "0.2.0" # Defaults to .Chart.AppVersion
-  serviceAccountName: ""
+    tag: "" # Defaults to .Chart.AppVersion
   nodeSelector: {}
+  serviceAccountName: ""
   # Example resources:
   # resources:
   #   requests:
@@ -19,19 +19,27 @@ logLevel: INFO
 
 # A GCS bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/dataset.json.
+# at /gcsDataset/gcs-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /gcsDataset/gcs-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 gcsPath: ""
 
 # An S3 bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/s3-dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/s3-dataset.json.
+# at /s3Dataset/s3-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /s3Dataset/s3-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 s3Path: ""
 
-# hfToken optionally creates a secret with the specified token.
-# Can be set using helm install --set hftoken=<token>
-hfToken: ""
+# Optional Token configuration for Hugging Face authentication.
+# hfSecret: Configures a pre-existing Kubernetes Secret.
+# hfToken: Creates a new kubernetes secret with the specified token.
+# If both specified, 'hfSecret' takes precedence over 'hfToken'.
+token:
+  hfSecret:
+    name: "" # The name of the existing Secret (e.g., 'my-hf-secret').
+    key: ""  # The key within the Secret that holds the token value (e.g., 'token' or 'hf-token').
+  hfToken: ""
 
 config:
   load:
@@ -44,8 +52,6 @@ config:
       duration: 30
     - rate: 500
       duration: 30
-    - rate: 700
-      duration: 30
     worker_max_concurrency: 1000
   api:
     type: completion
diff --git a/benchmarking/prefix-cache-aware/low-cache-values.yaml b/benchmarking/prefix-cache-aware/low-cache-values.yaml
@@ -2,9 +2,9 @@
 job:
   image:
     repository: quay.io/inference-perf/inference-perf
-    tag: "0.2.0" # Defaults to .Chart.AppVersion
-  serviceAccountName: ""
+    tag: "" # Defaults to .Chart.AppVersion
   nodeSelector: {}
+  serviceAccountName: ""
   # Example resources:
   # resources:
   #   requests:
@@ -19,19 +19,27 @@ logLevel: INFO
 
 # A GCS bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/dataset.json.
+# at /gcsDataset/gcs-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /gcsDataset/gcs-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 gcsPath: ""
 
 # An S3 bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/s3-dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/s3-dataset.json.
+# at /s3Dataset/s3-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /s3Dataset/s3-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 s3Path: ""
 
-# hfToken optionally creates a secret with the specified token.
-# Can be set using helm install --set hftoken=<token>
-hfToken: ""
+# Optional Token configuration for Hugging Face authentication.
+# hfSecret: Configures a pre-existing Kubernetes Secret.
+# hfToken: Creates a new kubernetes secret with the specified token.
+# If both specified, 'hfSecret' takes precedence over 'hfToken'.
+token:
+  hfSecret:
+    name: "" # The name of the existing Secret (e.g., 'my-hf-secret').
+    key: ""  # The key within the Secret that holds the token value (e.g., 'token' or 'hf-token').
+  hfToken: ""
 
 config:
   load:
@@ -44,8 +52,6 @@ config:
       duration: 30
     - rate: 500
       duration: 30
-    - rate: 700
-      duration: 30
     worker_max_concurrency: 1000
   api:
     type: completion
diff --git a/benchmarking/single-workload/decode-heavy-values.yaml b/benchmarking/single-workload/decode-heavy-values.yaml
@@ -2,9 +2,9 @@
 job:
   image:
     repository: quay.io/inference-perf/inference-perf
-    tag: "0.2.0" # Defaults to .Chart.AppVersion
-  serviceAccountName: ""
+    tag: "" # Defaults to .Chart.AppVersion
   nodeSelector: {}
+  serviceAccountName: ""
   # Example resources:
   # resources:
   #   requests:
@@ -19,19 +19,27 @@ logLevel: INFO
 
 # A GCS bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/dataset.json.
+# at /gcsDataset/gcs-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /gcsDataset/gcs-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 gcsPath: ""
 
 # An S3 bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/s3-dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/s3-dataset.json.
+# at /s3Dataset/s3-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /s3Dataset/s3-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 s3Path: ""
 
-# hfToken optionally creates a secret with the specified token.
-# Can be set using helm install --set hftoken=<token>
-hfToken: ""
+# Optional Token configuration for Hugging Face authentication.
+# hfSecret: Configures a pre-existing Kubernetes Secret.
+# hfToken: Creates a new kubernetes secret with the specified token.
+# If both specified, 'hfSecret' takes precedence over 'hfToken'.
+token:
+  hfSecret:
+    name: "" # The name of the existing Secret (e.g., 'my-hf-secret').
+    key: ""  # The key within the Secret that holds the token value (e.g., 'token' or 'hf-token').
+  hfToken: ""
 
 config:
   load:
diff --git a/benchmarking/single-workload/prefill-heavy-values.yaml b/benchmarking/single-workload/prefill-heavy-values.yaml
@@ -2,9 +2,9 @@
 job:
   image:
     repository: quay.io/inference-perf/inference-perf
-    tag: "0.2.0" # Defaults to .Chart.AppVersion
-  serviceAccountName: ""
+    tag: "" # Defaults to .Chart.AppVersion
   nodeSelector: {}
+  serviceAccountName: ""
   # Example resources:
   # resources:
   #   requests:
@@ -19,19 +19,27 @@ logLevel: INFO
 
 # A GCS bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/dataset.json.
+# at /gcsDataset/gcs-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /gcsDataset/gcs-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 gcsPath: ""
 
 # An S3 bucket path that points to the dataset file.
 # The file will be copied from this path to the local file system
-# at /dataset/s3-dataset.json for use during the run.
-# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /dataset/s3-dataset.json.
+# at /s3Dataset/s3-dataset.json for use during the run.
+# NOTE: For this dataset to be used, config.data.path must also be explicitly set to /s3Dataset/s3-dataset.json.
+# Format: bucket-name/folder/to/dataset/file
 s3Path: ""
 
-# hfToken optionally creates a secret with the specified token.
-# Can be set using helm install --set hftoken=<token>
-hfToken: ""
+# Optional Token configuration for Hugging Face authentication.
+# hfSecret: Configures a pre-existing Kubernetes Secret.
+# hfToken: Creates a new kubernetes secret with the specified token.
+# If both specified, 'hfSecret' takes precedence over 'hfToken'.
+token:
+  hfSecret:
+    name: "" # The name of the existing Secret (e.g., 'my-hf-secret').
+    key: ""  # The key within the Secret that holds the token value (e.g., 'token' or 'hf-token').
+  hfToken: ""
 
 config:
   load:
diff --git a/site-src/performance/benchmark/advanced-configs/decode-heavy-workload.md b/site-src/performance/benchmark/advanced-configs/decode-heavy-workload.md
diff --git a/site-src/performance/benchmark/advanced-configs/prefill-heavy-workload.md b/site-src/performance/benchmark/advanced-configs/prefill-heavy-workload.md
diff --git a/site-src/performance/benchmark/advanced-configs/prefix-cache-aware.md b/site-src/performance/benchmark/advanced-configs/prefix-cache-aware.md
diff --git a/site-src/performance/benchmark/index.md b/site-src/performance/benchmark/index.md
